A new mutation operation for faster convergence in genetic algorithm feature selection

Abstract

Feature selection is an important step in data classification because it has a high impact on classification accuracy. Feature selection using Genetic Algorithm (GA) is usually done in a wrapper method. The process is time consuming especially for large dimensional database. We propose a new mutation operation for faster feature selection by GA based on elitism of the allele. Normal elitism in GA preserves the most fit chromosomes which are evaluated using the fitness function. In the same way, the highest fit allele will be preserved and the fitness of the allele is evaluated based on the frequency of occurrences. The chromosome undergoing this mutation process will have a high if not the highest fitness because it is created based on a high fit allele. It will be the catalyst to increase the rate of convergence towards achieving an optimal features combination. Experiments for feature selection using this method are conducted using a database of tropical wood species which has a large variation of features. Results of the experiments show that a high accuracy is obtained for the recognition of the tropical wood species using the feature selection method. In addition, it has also been shown that the chromosomes created by the new mutation operation have high fitness and the rate of optimal convergence is improved substantially. The new mutation operation is not only useful for large database, but also can be used for small or medium sized database.